Intelligent monitoring system for coal conveying trestle belt

Abstract

The invention discloses an intelligent monitoring system for a coal conveying trestle belt, and the system comprises a data collection module which is used for collecting the image of the coal conveying belt and measuring the thickness distribution data of coal dust on the surface of the belt in real time, and obtaining a time sequence video stream and a coal dust thickness matrix of the coal conveying belt; the dynamic reference module is used for extracting a static belt structure according to the time sequence video stream to obtain an initial reference feature map as a belt reference image without coal dust interference; the coal dust sensing and coding module is used for inhibiting high-shielding area feature response according to the coal dust thickness matrix and coding the belt image of the current frame to obtain a feature map of the current frame; the dynamic difference analysis module is used for performing difference analysis according to the current frame feature map and the belt reference image to obtain an abnormal area; according to the method, high-definition imaging of the surface of the belt and stable extraction of abnormal characteristics can be realized in a complex environment of dynamic shielding of coal dust and nonlinear attenuation of optical signals.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and more particularly to an intelligent monitoring system for a coal conveying trestle belt. Background Art

[0002] Currently, high-definition video surveillance technology is widely used in safety monitoring systems for coal conveyor trestle belts. Industrial cameras are deployed to capture real-time footage of belt operation and use image processing algorithms (such as edge detection and template matching) to identify anomalies such as belt tears and deviation. This solution can meet basic detection needs in conventional industrial scenarios. For example, in low-dust environments, cameras can clearly capture cracks or edge deviations on the belt surface. Algorithms use pixel-level analysis to determine the location and severity of the anomaly, triggering an alarm signal.

[0003] However, in the actual operation environment of the coal conveyor trestle, the belt is exposed to high concentrations of coal dust (normal > 50mg / m 3 ) environment, coal dust particles form a non-uniform adhesion layer on the camera lens and the belt surface. This adhesion layer leads to two major problems: the scattering effect of coal dust particles on visible light and near-infrared bands (conventional camera working wavelengths) increases exponentially with the increase of accumulation thickness, and the image contrast of key features such as cracks and edge deformation on the belt surface will drop significantly. Traditional image enhancement algorithms (such as histogram equalization) can only linearly stretch the grayscale range and cannot restore the detailed features obscured by coal dust. At the same time, during the coal transportation process, the vibration of the belt and the impact of the coal flow cause the coal dust adhesion layer to continuously peel off and re-accumulate. In the video stream collected by the camera, the belt surface texture and coal dust distribution continue to change dynamically. Traditional matching algorithms based on fixed templates (such as SSIM structural similarity comparison) lack dynamic benchmark references, resulting in an extremely high misjudgment rate.

[0004] Therefore, how to achieve high-definition imaging of the belt surface and stable extraction of abnormal features in a complex environment with dynamic coal dust shielding and nonlinear attenuation of optical signals is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent monitoring system for coal conveyor trestle belts, which can achieve high-definition imaging of the belt surface and stable extraction of abnormal features in a complex environment with dynamic shielding of coal dust and nonlinear attenuation of optical signals.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] An intelligent monitoring system for a coal conveyor trestle belt, comprising:

[0008] The data acquisition module is used to collect images of the coal conveyor belt and measure the coal dust thickness distribution data on the belt surface in real time to obtain the time-series video stream of the coal conveyor belt and the coal dust thickness matrix;

[0009] A dynamic reference module is used to extract the static belt structure according to the time-series video stream and obtain an initial reference feature map as a belt reference image without coal dust interference;

[0010] a coal dust perception encoding module, configured to suppress the characteristic response of the high-shading area according to the coal dust thickness matrix and encode the belt image of the current frame to obtain a feature map of the current frame;

[0011] The dynamic difference analysis module is used to perform difference analysis based on the current frame feature map and the belt reference image to obtain an abnormal area.

[0012] Preferably, it also includes an abnormality analysis module, which is used to extract spatiotemporal features based on the abnormal area and analyze the spatiotemporal evolution feature graphs corresponding to different belt abnormalities.

[0013] Preferably, the abnormality analysis includes:

[0014] Stack the binary images of abnormal regions in multiple frames in time series into a three-dimensional tensor;

[0015] Parallel convolution branches are used to extract abnormal evolution features at different time scales, and pyramid pooling is performed on the spatial dimension to fuse local details with global morphological features to obtain the final spatiotemporal features.

[0016] Analyze the correlation between abnormal regions of adjacent frames through the self-attention mechanism to obtain a confidence matrix and weight the spatiotemporal features;

[0017] The preset anomaly pattern library is matched according to the weighted spatiotemporal features, and the anomaly type label and confidence score are output.

[0018] Preferably, when the dynamic reference module separates the static belt structure from the dynamic coal dust interference through background modeling, an adaptive update strategy based on the coal dust distribution is adopted: the coal dust covered area is dynamically masked, and the background model is updated only for the coal dust-free or low-shielding areas to retain the integrity of the belt structure.

[0019] Preferably, the adaptive update strategy of the dynamic reference module includes the following steps:

[0020] Generate a dynamic mask based on the coal dust thickness matrix, mark the area where the coal dust thickness exceeds the preset threshold as a high-occlusion area, and update the background sample only for pixels in the low-occlusion area of ​​the non-mask;

[0021] The geometric characteristics of the static belt structure are periodically calibrated in combination with the belt load variation data.

[0022] Preferably, the adaptive update strategy of the dynamic reference module also includes: tracking the drifting trajectory of coal dust in the high-shading area by the optical flow method, and if the movement speed of the coal dust exceeds a preset threshold, eliminating the dynamic interference signal of the area in the reference feature map.

[0023] Preferably, in the coal dust perception coding module, a dynamic weight map is generated by using a coal dust thickness matrix to locally suppress the feature response of the high-shading area, while interpolating and compensating the features of the adjacent low-shading area to ensure the continuity of the belt texture.

[0024] Preferably, the feature suppression and compensation steps of the coal dust perception coding module include:

[0025] The coal dust thickness matrix is ​​input into the convolution layer to generate a dynamic weight map, and the weight value is inversely proportional to the coal dust thickness;

[0026] Perform weighted processing on the feature map output by the encoder to suppress the feature response of the highly occluded area;

[0027] Locally interpolate the adjacent features of the suppressed area and use the feature mean of the adjacent low-occlusion area to restore texture continuity;

[0028] The original feature map is fused with the compensated feature map through skip connections to preserve the global context information.

[0029] Preferably, in the dynamic difference analysis module, the calculation of the difference matrix introduces the coal dust thickness weight, and the difference value of the high-shading area is dynamically attenuated to avoid misjudgment caused by coal dust drift.

[0030] Preferably, the difference value attenuation of the dynamic difference analysis module is achieved by the following steps:

[0031] Normalize the coal dust thickness matrix into an attenuation coefficient matrix, the greater the thickness, the higher the attenuation coefficient;

[0032] Perform pixel-by-pixel attenuation on the difference matrix: difference value = original difference value × (1-attenuation coefficient);

[0033] Morphological filtering is performed on the attenuated difference matrix to retain the continuous structural features of the belt edge and cracks;

[0034] Isolated noise points are eliminated through connected domain analysis, and a binary map of abnormal areas is output.

[0035] Through the above technical solutions, it can be seen that compared with the prior art, the present invention discloses an intelligent monitoring system for coal conveyor trestle belts. The dynamic reference module separates the static belt structure from the time-series video stream and generates a reference image without coal dust interference. By adaptively filtering out the instantaneous interference of dynamic coal dust, the module ensures the stable expression of key geometric features such as the belt edge and bracket, avoiding the blurring or deformation of the reference image caused by the drifting of coal dust. Even in areas covered by high-concentration coal dust, the spatial topological relationship of the belt infrastructure is still accurately preserved, providing a reliable reference for subsequent difference analysis. The coal dust perception coding module dynamically adjusts the feature response intensity based on the coal dust thickness matrix: it suppresses features in high-shading areas (such as coal dust accumulation areas) to block the propagation of coal dust noise to the deep network; at the same time, it enhances the texture details (such as cracks and foreign matter) in low-shading areas through the coding process. This mechanism realizes the preliminary separation of noise and effective signals in the feature extraction stage, so that the current frame feature map contains both local details under coal dust interference and maintains the semantic consistency of the overall belt structure. The dynamic difference analysis module directly focuses on the deviation of the belt surface from the static reference by comparing the current frame feature map with the reference image. Because the reference image eliminates the dynamic interference of coal dust, and the current frame feature map suppresses noise in highly obscured areas through encoding, the difference matrix between the two can accurately reflect the difference in optical attenuation characteristics between real anomalies (such as tears and foreign objects) and coal dust-obscured areas. This module mechanically distinguishes between dynamic interference (instantaneous differences caused by coal dust flutter) and static anomalies (persistent structural defects), thereby achieving robust detection of anomaly areas in complex optical attenuation environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0037] Figure 1 The accompanying drawing is a schematic structural diagram of an intelligent monitoring system for a coal conveyor trestle belt provided by the present invention.

[0038] Figure 2 Schematic diagram of the data processing flow of the dynamic benchmark module in an embodiment of the present invention.

[0039] Figure 3 This is a flow chart of coal dust sensing coding in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] like Figure 1 The embodiment of the present invention discloses an intelligent monitoring system for a coal conveyor trestle belt, comprising:

[0042] The data acquisition module is used to collect images of the coal conveyor belt and measure the coal dust thickness distribution data on the belt surface in real time, thereby obtaining a time-series video stream of the coal conveyor belt and a coal dust thickness matrix;

[0043] A dynamic reference module is used to extract the static belt structure based on the time-series video stream and obtain an initial reference feature map as a belt reference image without coal dust interference;

[0044] A coal dust perception and encoding module is used to suppress the characteristic response of the high-shading area according to the coal dust thickness matrix and encode the belt image of the current frame to obtain a feature map of the current frame;

[0045] The dynamic difference analysis module is used to perform difference analysis based on the current frame feature map and the belt reference image to obtain the abnormal area.

[0046] In this embodiment, the data acquisition module combines a polarization-compensated camera with a laser scattering sensor to simultaneously capture a high-definition video stream and a coal dust thickness matrix in a dusty environment. Near-infrared polarized light penetrates the coal dust layer and suppresses scattering noise, while laser scattering directly quantifies the degree of obscuration. These two complementary methods overcome the image blur and feature loss caused by dust obscuration in traditional single-camera systems. The dynamic reference module extracts the static belt structure from the video stream to generate a reference image. Using background modeling, it separates the belt structure from the dynamic coal dust. Using optical flow, it tracks the coal dust's motion trajectory and dynamically removes interfering areas. Load data is also incorporated to calibrate belt deformation, ensuring the reference consistently reflects the true structure and avoiding misjudgments caused by traditional static references due to coal dust movement or load fluctuations. The coal dust perception and encoding module generates dynamic weights based on the coal dust thickness matrix to suppress feature responses in highly obscured areas. Using neighborhood interpolation, it compensates for obscured belt texture, eliminating dust movement noise while preserving key features such as cracks and edges. This addresses the missed detections often associated with traditional algorithms due to feature loss in obscured areas. The dynamic difference analysis module introduces an attenuation coefficient driven by coal dust thickness when calculating inter-frame differences, reducing the reliability of differences in highly shielded areas. It also removes isolated noise through morphological filtering and focuses on continuous abnormal areas (such as strip deviation and radial cracks), making the detection results resistant to instantaneous interference and maintaining high-precision positioning.

[0047] In order to further implement the above technical solution, the anomaly analysis module is used to extract spatiotemporal features based on abnormal areas and analyze the spatiotemporal evolution feature maps corresponding to different belt anomalies. The steps include: stacking the binary images of multiple consecutive frames of abnormal areas into a three-dimensional tensor according to the time series; extracting abnormal evolution features at different time scales through parallel convolution branches, and performing pyramid pooling on the spatial dimensions to fuse local details with global morphological features to obtain the final spatiotemporal features; analyzing the correlation between abnormal areas of adjacent frames through the self-attention mechanism to obtain a confidence matrix and weight the spatiotemporal features; matching the preset abnormal pattern library according to the weighted spatiotemporal features, and outputting the abnormal type label and confidence score.

[0048] Specifically, first, the binary images of multiple consecutive frames of abnormal areas are stacked into a three-dimensional tensor in time series to construct a spatiotemporal data carrier. This tensor retains the dynamic change information of the abnormal area in the time dimension (such as the deviation trend and the direction of crack diffusion), and maps the position and morphological distribution of the abnormal area in the spatial dimension (such as strip-shaped and radial). Then, the abnormal evolution characteristics of different time scales are extracted through parallel convolution branches - the short-term branch captures the inter-frame mutation (such as the instantaneous appearance of foreign objects), the medium-time branch models the trend offset (such as the gradual diffusion of deviation), and the long-time branch analyzes the cumulative effect (such as the slow extension of cracks). At the same time, the spatial dimension is pyramid pooled to fuse local details (crack edge sharpness) and global morphology (overall deformation trend of the belt) to ensure that the feature expression takes into account both micro and macro information.

[0049] Furthermore, a self-attention mechanism is used to enhance temporal correlation: a query vector (Query), a key vector (Key), and a value vector (Value) are generated for each abnormal region in the three-dimensional tensor. Similarity weights are then calculated for abnormal regions at the same location in adjacent frames. If a region is consistently active across multiple frames (e.g., ≥5 frames), its confidence weight is increased; if it only appears in a single frame, it is considered transient noise and its weight is reduced. This mechanism effectively distinguishes true anomalies (continuously evolving) from transient interference (random noise).

[0050] Finally, the weighted spatiotemporal features are matched against a pre-defined library of anomaly patterns: deviation patterns correspond to continuous linear deviations in edge regions, tear patterns match radial crack propagation trajectories, and foreign object intrusion patterns identify random jumps in isolated regions. Based on the matching results, an anomaly type label and confidence score are output, triggering a graded alarm. This process, through spatiotemporal modeling and pattern matching driven by physical laws, addresses the inability of traditional threshold methods to distinguish anomaly types and their susceptibility to transient interference, thereby improving detection interpretability and reliability.

[0051] Among them, the construction method of the exception pattern library is:

[0052] Deviation mode: Extracts the progressive deviation features of the belt edge over multiple consecutive frames, showing a band-like diffusion in space and a linear increase in the deviation in time;

[0053] Tear mode: Detects the extension direction and speed of cracks on the belt surface. The cracks are distributed radially in space and increase exponentially in length over time.

[0054] Foreign body intrusion pattern: Identify isolated abnormal areas of non-belt structures with abrupt contours in space and random changes in regional position in time.

[0055] like Figure 2 ,In order to further implement the above technical scheme, the ,dynamic benchmark module uses an adaptive update strategy based on ,coal dust distribution when separating the static belt structure and the dynamic coal dust ,interference through background modeling: the coal dust covered area is ,dynamically masked, and the background model is only updated for ,coal dust free or low shielding areas to preserve the ,integrity of the belt structure.

[0056] The adaptive update strategy of the dynamic reference module includes the following steps: generating a dynamic mask based on the coal dust thickness matrix, marking the area where the coal dust thickness exceeds a preset threshold as a high-shading area, so that background samples are updated only for pixels in the non-masked low-shading area; tracking the drifting trajectory of coal dust in the high-shading area by the optical flow method. If the coal dust movement speed exceeds the preset threshold, the dynamic interference signal of the area is eliminated from the reference feature map; and periodically calibrating the geometric characteristics of the static belt structure based on the belt load change data.

[0057] For example, a laser scanner generates a real-time coal dust distribution matrix on the belt surface, detecting that the coal dust accumulation on the right half of the belt reaches a thickness of 5-8mm (with a preset threshold of 3mm), while the coal dust on the left half is thinner (0-2mm). The system then automatically marks the right half as a "highly obscured area" and generates a dynamic mask to cover this area, allowing only the low-obscured areas on the left half to participate in the background model update. This strategy avoids noise contamination in the coal dust accumulation area and preserves the original features of static structures such as the belt edge and bracket.

[0058] In the monitoring image, coal dust in the high-shading area is floating due to the influence of airflow. Using the optical flow method, the system captures a group of coal dust moving toward the upper right at a speed of 15 pixels / frame (the preset speed threshold is 10 pixels / frame), and immediately marks the pixel area covered by the motion trajectory as a dynamic interference signal, and removes the pixel data at these locations from the feature map of the current frame. At the same time, fixed foreign objects (such as metal fragments) are detected on the belt surface in the low-shading area. Since the background model of this area is not contaminated, the system accurately identifies the location of the foreign object and triggers an alarm.

[0059] When the belt load increases from 500 tons to 800 tons, the weighing sensor synchronously transmits load data. Based on the preset belt deformation model, the system predicts that the sagging of the middle belt under the current load will increase by 12%, and then dynamically compensates and calibrates the geometric parameters of the belt in the baseline model (such as edge curvature and bracket spacing). During the calibration process, the system preferentially uses the undeformed structural features in the low-shading area as a reference point to ensure that the corrected background model is consistent with the actual belt shape. This process is performed periodically every 30 seconds to continuously adapt to the slow deformation of the belt caused by load changes.

[0060] In this embodiment, in a coal mine transport belt monitoring system, the present invention dynamically adjusts the monitoring system's calibration of the belt's physical shape (e.g., curvature, edge position, etc.) based on changes in the actual weight of coal carried by the belt, thereby preventing belt deformation caused by load changes from being misjudged as abnormalities (e.g., deviation or breakage). The following are several specific examples:

[0061] Dynamic belt sag compensation: While the belt is straight when unloaded, it sags in the middle, forming an arc when fully loaded (for example, sag increases by 15 cm). Calibration logic: The load cell detects the current load as 800 tons (preset full load is 1000 tons). The system calculates the expected sag based on the deformation model (for example, sag increases by 3 cm for every 100-ton increase in load). The baseline in the middle of the belt is adjusted from a straight line to a curved line offset 9 cm downward. In subsequent testing, the system uses this adjusted curve as a reference to determine if abnormal sag has occurred.

[0062] Correction for lateral deviation of the belt edge position: When the load increases, the belt may expand slightly to the sides due to gravity (for example, 2 cm on each side). Calibration logic: Load data triggers the calibration cycle, and the system obtains the actual position of the current edge (for example, the left edge shifts 1.8 cm to the right, and the right edge shifts 2.1 cm to the left); the "standard edge line" for visual inspection is adjusted from fixed coordinates to a dynamic range (for example, the allowed position range of the left edge changes from X = 100 ± 1 cm to X = 101.8 ± 1 cm); if the edge is detected outside the corrected range, it is judged as deviation.

[0063] Belt bracket spacing deformation compensation: When the load is too heavy, the belt between the two fixed brackets may stretch, resulting in a visual "increased bracket spacing" (for example, the actual bracket spacing is 5 meters, but the stretched image shows 5.2 meters). Calibration logic: Calculate the belt stretch ratio based on load data (such as 4% stretching at a load of 800 tons); dynamically correct the original bracket spacing value of 5 meters to 5.2 meters; if the spacing is detected to exceed ±5% of the corrected value (for example, >5.46 meters), a bracket loosening or breakage alarm is triggered.

[0064] like Figure 3,In order to further implement the above technical scheme, in the coal dust ,perception encoding module, a dynamic weight map is generated through the coal dust thickness ,matrix to locally suppress the feature response of the high-shading area, ,while interpolating and compensating the features of the adjacent low-shading ,areas to ensure the continuity of the belt texture.

[0065] The steps of feature suppression and compensation of the coal dust perception encoding module include: inputting the coal dust thickness matrix into the convolutional layer to generate a dynamic weight map, and the weight value is inversely proportional to the coal dust thickness; weighting the feature map output by the encoder to suppress the feature response of the high-occlusion area; locally interpolating the adjacent features of the suppressed area, and using the feature mean of the adjacent low-occlusion area to restore texture continuity; and fusing the original feature map with the compensated feature map through jump connections to retain global context information.

[0066] For example,

[0067] The system first uses a laser scanner to obtain the coal dust thickness distribution matrix on the belt surface. For example, the coal dust accumulation thickness in the middle section of the belt (coordinate range X = 100-200, Y = 50-150) reaches 5-8 mm, while the thickness of the edge areas is only 0.2-1 mm. This thickness data is input into a single-channel 1×1 convolution layer. The convolution kernel parameters are designed as negative correlation functions. For example, the weight calculation uses the formula (where d is the thickness of the coal dust) and uses a sigmoid function to compress the weights to between 0 and 1. For areas with a thickness of 8 mm, the weight is reduced to approximately 0.15, while for areas with a thickness of 0.5 mm, the weight remains at 0.89. The result is a dynamic weight map that is inversely proportional to the thickness of the coal dust, with highly obscured areas appearing darker (lower weight) and less obscured areas appearing lighter (higher weight).

[0068] Next, in the original feature map extracted by the encoder (such as ResNet-34), the texture features of the middle section of the belt are blurred due to coal dust coverage. For example, the activation value of a key feature channel representing the belt grid is 0.9 in the normal area, but is suppressed to 0.3 in the middle section. The system multiplies the dynamic weight map with the original feature map channel by channel, further reducing the activation value of the highly occluded area in the middle section. For example, when the weight of a pixel point is 0.2, the feature values ​​of all channels at that point are scaled to one-fifth of the original value, and the above-mentioned key feature is reduced from 0.3 to 0.06. This operation significantly weakens the noise signal in the coal dust interference area, but it also causes local loss of texture information.

[0069] To solve this problem, the system uses local interpolation compensation for each suppressed pixel (such as an area with a weight less than 0.3): with a high-shading point (X=120, Y=80) as the center, a 5×5 neighborhood window is defined, and pixels with weights higher than 0.7 (such as the 8 adjacent low-shading points) are screened, and the feature mean of these points is calculated. For example, the activation value of the high-shading point in a certain channel was originally suppressed to 0.06. By taking the average of the 8 surrounding points (assuming that the feature values ​​of these points are 0.55-0.65), the feature of the point is restored to 0.58 after interpolation, which is close to 0.65 in the normal area. This process allows the middle area covered by coal dust to reconstruct a coherent belt grid texture. For example, the originally broken longitudinal stripes are aligned with the unshaded areas on both sides after interpolation.

[0070] Finally, the system fuses the compensated feature map with the original encoder output via skip connections. Specifically, the compensated feature map (256×256×64 channels) and the original feature map (also 256×256×64 channels) are concatenated along the channel dimension to form a 256×256×128 fused feature map, which is then compressed back to 64 channels via a 1×1 convolution. During this process, the convolution kernel automatically learns the weight distribution between the two types of features. For example, in low-occlusion areas, the original features retain 70% of the weight to preserve details (such as the high response value of 0.88 for a 0.5 mm crack on the right edge), while in high-occlusion areas, the compensated features take up 85% of the weight to enhance the repaired structure (such as the 0.72 response value for a longitudinal tear in the middle section). The fused feature map preserves subtle anomalies unaffected by coal dust while restoring the continuous texture in the occluded areas. When ultimately transmitted to the downstream detection module, it can successfully identify the longitudinal tear in the middle section obscured by coal dust (feature response exceeds a threshold of 0.6) and the undisturbed transverse cracks on both sides.

[0071] In order to further implement the above technical solution, the calculation of the difference matrix introduces a coal dust thickness weight, and dynamically attenuates the difference value in the highly shielded area to avoid misjudgment caused by coal dust fluttering; the difference value attenuation is achieved through the following steps: the coal dust thickness matrix is ​​normalized into an attenuation coefficient matrix, and the greater the thickness, the higher the attenuation coefficient; the difference matrix is ​​attenuated pixel by pixel: difference value = original difference value × (1-attenuation coefficient); the attenuated difference matrix is ​​morphologically filtered to retain the continuous structural characteristics of the belt edge and cracks; isolated noise points are eliminated through connected domain analysis, and a binary image of the abnormal area is output.

[0072] In this embodiment, the system first uses a laser scanner to obtain the coal dust thickness distribution data on the current belt surface and generates a coal dust thickness matrix. In this matrix, the value of the middle section of the belt is significantly higher than that of the edge areas on both sides due to the thicker coal dust accumulation. In order to convert the thickness data into an attenuation coefficient, the system normalizes the thickness matrix - the maximum thickness value is mapped to an attenuation coefficient of 1.0, the minimum thickness value is mapped to 0, and other positions are calculated according to a linear scale. For example, if the coal dust thickness in a certain area is 80% of the maximum value, its attenuation coefficient is 0.8. In the normalized attenuation coefficient matrix, high-shading areas show high attenuation values ​​close to 1, while low-shading areas are close to 0.

[0073] Next, the system compares the current frame image with the background model at the pixel level to generate the original difference matrix. In the original difference matrix, the coal dust floating area produces scattered high difference values ​​due to dynamic interference, while the real crack area presents a continuous high difference band. At this time, the system performs a pixel-by-pixel attenuation operation on the difference matrix: the difference value of each pixel is multiplied by (1-the attenuation coefficient of the corresponding position). For example, if the difference value of the high-shading area is originally 0.9, it may be reduced to 0.18 (0.9×(1-0.8)) after attenuation, while the difference value of 0.9 in the low-shading area is only slightly reduced to 0.81 (0.9×(1-0.1)). This operation significantly weakens the dynamic interference of coal dust, but may cause the real crack area to break (for example, when the crack passes through the high-shading area, some pixels are over-attenuated).

[0074] To repair fractured structures, the system performs a morphological closing operation on the attenuated difference matrix: dilating the matrix using a 3×3 rectangular kernel followed by erosion. This operation connects adjacent high-difference regions. For example, a crack split into two segments by a highly obscured area can be filled in after the closing operation, forming a continuous region. Subsequently, an opening operation (erosion followed by dilation) eliminates isolated noise points. For example, random noise points return their difference values ​​to zero after erosion, and dilation still cannot restore them. However, the true cracks are preserved due to their strong continuity.

[0075] Finally, the system performs a connected domain analysis on the processed difference matrix. All high-difference areas are marked using the 8-neighborhood connectivity rule, and the area and average difference value of each connected area are calculated. An area threshold (such as a minimum of 20 pixels) is set to filter out tiny noises, and only areas exceeding the threshold are retained as abnormal outputs. For example, sporadic points caused by coal dust drift are eliminated due to insufficient area, while the long strip-shaped connected domain formed by real cracks is completely retained. The output results mark the abnormal location in the form of a binary map, directly guiding on-site personnel for maintenance.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0077] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent monitoring system for a coal conveyor trestle belt, characterized in that: include: The data acquisition module is used to collect images of the coal conveyor belt and measure the coal dust thickness distribution data on the belt surface in real time to obtain the time-series video stream of the coal conveyor belt and the coal dust thickness matrix; A dynamic reference module is used to extract the static belt structure according to the time-series video stream and obtain an initial reference feature map as a belt reference image without coal dust interference; a coal dust perception encoding module, configured to suppress the characteristic response of the high-shading area according to the coal dust thickness matrix and encode the belt image of the current frame to obtain a feature map of the current frame; The dynamic difference analysis module is used to perform difference analysis based on the current frame feature map and the belt reference image to obtain an abnormal area.

2. The intelligent monitoring system for a coal conveyor trestle belt according to claim 1 is characterized in that: It also includes an abnormality analysis module, which is used to extract spatiotemporal features based on the abnormal area and analyze the spatiotemporal evolution feature maps corresponding to different belt abnormalities.

3. The intelligent monitoring system for coal conveyor trestle belt according to claim 2 is characterized in that: The abnormality analysis includes: Stack the binary images of abnormal regions in multiple frames in time series into a three-dimensional tensor; Parallel convolution branches are used to extract abnormal evolution features at different time scales, and pyramid pooling is performed on the spatial dimension to fuse local details with global morphological features to obtain the final spatiotemporal features. Analyze the correlation between abnormal regions of adjacent frames through the self-attention mechanism to obtain a confidence matrix and weight the spatiotemporal features; The preset anomaly pattern library is matched according to the weighted spatiotemporal features, and the anomaly type label and confidence score are output.

4. The intelligent monitoring system for a coal conveyor trestle belt according to claim 1, characterized in that: When the dynamic reference module separates the static belt structure from the dynamic coal dust interference through background modeling, it adopts an adaptive update strategy based on the coal dust distribution: the coal dust covered area is dynamically masked, and the background model is updated only for the coal dust-free or low-shielding areas to preserve the integrity of the belt structure.

5. The intelligent monitoring system for coal conveyor trestle belt according to claim 4 is characterized in that: The adaptive update strategy of the dynamic benchmark module includes the following steps: Generate a dynamic mask based on the coal dust thickness matrix, mark the area where the coal dust thickness exceeds the preset threshold as a high-occlusion area, and update the background sample only for pixels in the low-occlusion area of ​​the non-mask; The geometric characteristics of the static belt structure are periodically calibrated in combination with the belt load variation data.

6. The intelligent monitoring system for coal conveyor trestle belt according to claim 5, characterized in that: The adaptive update strategy of the dynamic reference module also includes: tracking the drifting trajectory of coal dust in the high-shading area through the optical flow method, and if the coal dust movement speed exceeds a preset threshold, eliminating the dynamic interference signal of the area in the reference feature map.

7. The intelligent monitoring system for coal conveyor trestle belt according to claim 1, characterized in that: In the coal dust perception coding module, a dynamic weight map is generated through the coal dust thickness matrix to locally suppress the feature response of the high-shading area, while interpolating and compensating the features of the adjacent low-shading area to ensure the continuity of the belt texture.

8. The intelligent monitoring system for coal conveyor trestle belt according to claim 7, characterized in that: The steps of suppressing and compensating the features of the coal dust sensing coding module include: The coal dust thickness matrix is ​​input into the convolution layer to generate a dynamic weight map, and the weight value is inversely proportional to the coal dust thickness; Perform weighted processing on the feature map output by the encoder to suppress the feature response of the highly occluded area; Locally interpolate the adjacent features of the suppressed area and use the feature mean of the adjacent low-occlusion area to restore texture continuity; The original feature map is fused with the compensated feature map through skip connections to preserve the global context information.

9. The intelligent monitoring system for coal conveyor trestle belt according to claim 1, characterized in that: In the dynamic difference analysis module, the calculation of the difference matrix introduces the coal dust thickness weight, and dynamically attenuates the difference value of the high-shading area to avoid misjudgment caused by coal dust drift.

10. The intelligent monitoring system for coal conveyor trestle belt according to claim 9, characterized in that: The difference value attenuation of the dynamic difference analysis module is achieved by the following steps: Normalize the coal dust thickness matrix into an attenuation coefficient matrix, the greater the thickness, the higher the attenuation coefficient; Perform pixel-by-pixel attenuation on the difference matrix: difference value = original difference value × (1-attenuation coefficient); Morphological filtering is performed on the attenuated difference matrix to retain the continuous structural features of the belt edge and cracks; Isolated noise points are eliminated through connected domain analysis, and a binary map of abnormal areas is output.

Images